摘要
针对图像在动态阈值选取难的问题,通过比较全局阈值和局部阈值优缺点,选用贝叶斯阈值估计和迭代加权的方法对图像进行二值化分割,建立基于贝叶斯线性回归模型对检测到的阈值进行分析,通过图像增强,建立目标与非目标区域,分别计算各个区域的先验概率,使用贝叶斯估计模型求得似然函数的极小值即为后验概率,通过此模型对125组阈值样本进行分类,对异常阈值的判断率为14.4%,选取后的阈值更为精确。本文方法,既能有效的提取目标特征,较好的去除背景,又能够保留目标图像的细节。
Image dynamic threshold selection is a difficult problem,by comparing advantages and disadvantages of the global threshold and the local threshold.Bayesian threshold estimation method and iterative weighted image binarization segmentation are selected.Based on the analysis of the threshold detected from the Bayesian linear regression model,the threshold value can be more precise which is obtained from the image enhancement and Bayesian estimation.Classify by the fact that this model is in progress to 125 set of threshold value sample books,the judgement rate to abnormal threshold value is 14.4%.The results indicated that that this method mentioned in this essay can not only target feature extraction effectively,remove the background more evidently,but also retain the target image details.
出处
《价值工程》
2012年第17期151-152,共2页
Value Engineering
关键词
图像分割
贝叶斯模型
全局法
局部法
segmentation
bayesian model
global method
local approach
